fregrid and non-regriddable input diagnostics

[ceblanton]

Can fregrid detect whether it can regrid a dataset satisfactily? If not, how can the workflow best communicate this to fregrid?

Before CMIP6 there was little interest or capability to regrid the ocean or ice tripolar grid to lat/lon. During CMIP6 fregrid and frepp updates were made to do this.

One of the new problems was that some tripolar variables couldn't be regridded. Mostly, there were vector fields that couldn't be regridded while conserving global sums (we believe) and some other scalar fields that cannot be regridded to valid output (mostly grid related). When attempting to regrid these non-regriddable variables, fregrid will sometimes give an error (for vectors; due to dimension name conflicts) but for scalar fields that can't be regridded it will go ahead and try, producing junk output.

The solution we went with was overloading the "interp_method" variable attribute to be set to "NONE" if it can't be regridded. (Normally it is set to the default regridding method, e.g. conserve_order1 or conserve_order2).

e.g. This commit in SIS2 made some of these changes to prevent FRE and fregrid from regridding fields that shouldn't be regridded:

NOAA-GFDL/SIS2@916cb4a

This type of fix is disagreeable as it is a fregrid setting embedded in the SIS2 model code, and the SIS2 developers would prefer to not make such changes in the future.

If fregrid can't determine whether a given diagnostic can be regridded with scientifically valid output, then users need to make this assessment and communicate it to the workflow. Setting this variable "interp_method" attribute is the only method to do this currently. Is there a better option?

There is another mechanism to do this identification of non-regriddable variables within a fre.properties setting, but this is a solution for the current workflow only (Bronx).